Obtaining value from the human genome: a challenge for the pharmaceutical industry

perspective REVIEWS

DDT Vol. 3, No. 4 April 1998 Copyright © Elsevier Science Ltd. All rights reserved. 1359-6446/98/$19.00. PII: S1359-6446(97)01162-8 179

In a relatively short period of time, the sequence of the

entire human genome and the identity of the genes

contained within it will be determined. Among this vast

array of genes are the pharmaceutical targets of the

future. To benefit fully from this explosion of data, the

pharmaceutical industry must establish high-throughput

target identification and validation processes to feed

their development pipelines.

In the past few years, the pharmaceutical industry has been

confronted by a series of challenges arising from the

ever-increasing advances in technology. In a competitive

industry, any technology that allows higher productivity

needs to be adopted and effectively integrated into the pro-

duction process as rapidly as possible. Within the pharma-

ceutical industry, this has meant the integration of high-

throughput screening and combinatorial chemistry into the

discovery process, together with the ever-increasing use of

complex databases to organize and store the vast quantities

of data such technologies create. The Human Genome Project,

initiated largely as an academic exercise less than a decade

ago, is adding another dimension to the equation. In order

to utilize this new and immensely valuable database fully (it

will soon contain the DNA and protein sequences of every

drug target the industry will ever work on), the pharmaceu-

tical industry must establish efficient processes to determine

which genes or gene products have real therapeutic value.

Although the Human Genome Project is officially only six

years old, we are already being informed in editorials

and conference announcements that we are entering

the ‘postgenomics’ era. For the pharmaceutical industry this

means that attention is turning towards the biological charac-

terization of the thousands of new genes that are catalogued

in public and private databases. Furthermore, in most cases,

the genes or gene products are being evaluated as the targets

for small-molecule therapeutics. This review describes how

the tools of bioinformatics, genetics, molecular biology and

pharmacology may be applied to triage the data coming

from the Human Genome Project and other genomic pro-

grams efficiently, and thus identify the most likely candidate

genes for pharmaceutical development. An attempt has been

made to describe the process as generically as possible,

therefore, some of the subtleties of analysis that might be

applied in a specific disease area have been ignored.

However, an overview of both the process and the tech-

nologies necessary to ‘cherry pick’ the most valuable thera-

peutic targets in a timely manner is provided.

Analysis of the human genome and gene expressionThe human genome comprises about 3 3 109 bp of DNA of

which ~5% codes for genes. Various estimates have been

given regarding the number of genes encoded by the

human genome, but it is generally thought that there are

about 50,000–100,000 (Refs 1,2). It is anticipated that the

entire human genome will be sequenced by the year 2005

(Ref. 3). In parallel with these efforts, several academic and

commercial centres are carrying out high-throughput

Obtaining value from the humangenome: a challenge for thepharmaceutical industryPaul Spence

Paul Spence, CNS Disorders Division, Wyeth-Ayerst Research, CN8000 Princeton, NJ 08543-8000, USA. tel: +1 732 274 4027, fax: +1 732 274 4020, e-mail: [email protected]

sequencing of expressed genes and generating databases of

expressed sequence tags (ESTs). These databases are of par-

ticular value to pharmaceutical companies because they

provide sequence data that are naturally filtered, albeit

coarsely, for pharmaceutical evaluation (the databases only

contain sequences from exons of expressed genes). They

also allow comparisons of genes between species and the

discovery of new gene families involved in human disease4.

Although valuable (~50% of human genes have been sampled

as ESTs; Ref. 5), EST databases are presently poorly anno-

tated and largely provide a starting point for further work.

The haploid human genome is divided among 23 chro-

mosomes, each gene having a specific locus on its specific

chromosome. A combination of the technologies and infor-

mation that have been developed for the physical mapping

of genes and family studies of inheritance has led to the

identification of several disease genes over the past few

years; for example, some genes have now been associated

with Alzheimer’s disease6.

Recently, the work carried out by the groups mapping

genes and identifying genes by EST sequencing was com-

bined so that a gene map of increasing resolution could be

constructed before the sequence of the entire human

genome became available3. Such a map will greatly facilitate

the identification of disease genes by the positional candi-

date approach, where large genomic regions identified in

complex-trait studies can be rapidly analyzed based on their

known gene content7.

In addition to the above efforts, many pharmaceutical and

biotechnology companies are generating large databases

of ESTs focused around specific tissue types, diseased

and normal tissues and tissues from various stages of embry-

onic development. However, as the genomic databases are

increasing in size at a near exponential rate, the pharma-

ceutical industry is increasingly turning its attention to the

elucidation of gene function, with the concern that the

functional evaluation of genes is not yet a high-throughput

science.

Focusing the search for disease genesWith the requirement to select the best targets from a

pool of 100,000 candidates, it makes sense to reduce this

REVIEWS perspective

180 DDT Vol. 3, No. 4 April 1998

Two methods are currently in widespread use. The firstinvolves the arraying of cDNAs, normally onto glass slides, followed by hybridization of fluorescently labelled probesderived from reverse-transcribed mRNA from the tissue ofinterest (Figure 2). Because of the small format and high den-sity of cDNAs on the solid phase, it is claimed that such arrayscan give both sensitive and quantitative data on changes inmRNA expression levels for many genes simultaneously37,38.Such arrays can be used, for example, to compare the geneexpression profiles of normal and diseased tissues.Differential display techniques do not rely on the availabilityof cDNAs but use quantitative polymerase chain reaction(PCR) amplification and primer combinations that can effec-tively amplify all sequences within a population of mRNAs(Ref. 39). These methods can be used to clone novel genesby virtue of their differential expression patterns in differenttissues, or in the same tissue under different circumstances.As the technologies for transcriptional profiling becomemore sophisticated – allowing highly quantitative analysis,accommodating a higher sample throughput and more-complete coverage of the entire expression repertoire ofcells and tissues – it will soon be possible to generate com-plete transcriptional fingerprints40,41. Such information mightenable the analysis of signal transduction pathways that aremodulated in diseases such as cancer42, as well as the‘whole-genome’ activity of pharmaceuticals. It is likely thattranscriptional profiling will be used less to isolate individual

genes through their change in expression, but more toidentify patterns of transcriptional modulation leading to amore complete understanding of the biology of diseaseprocesses.Database mining is required at all stages of the gene dis-covery and validation process. It can be used for novelgene discovery, by homology searching of databases forsequences representing known gene sequences or sub-sequences43. Although often seen as a ‘fast track’ way ofidentifying new genes, care needs to be taken in designingsearch strategies and interpreting the results44. The reason-able assumption is that genes showing sequence homologyare likely to show functional homology45. In addition, hom-ology searching is capable of detecting relationships amonggenes that are separated by billions of years of evolution.This is well illustrated by the use of bioinformatics to charac-terize the function of the product of the ataxia telangi-ectasia gene in silico by its homology to the yeast phos-phatidylinositol 3-kinase, in addition to FK506- and rapamycin-binding proteins46,47. Although not an experimental methodfor focusing gene discovery, database mining does allow theluxury of choosing the ‘type’ of gene one would like to study.Many valuable pharmaceutical targets such as the G protein-coupled receptors are present in the human genome aslarge superfamilies, and a great deal is known about thesequence motifs present in these proteins such that focuseddatabase mining can be used to identify novel members48.

Box 1. Transcriptional profiling and database mining

candidate pool size as soon as possible. The pharmaceutical

industry, as well as many academic groups, has gone about

this in several ways (see Box 1 and Figures 1–3). As will

become evident, each approach has its advantages and

disadvantages, which will require different emphases to be

placed on the evaluation process (Box 2).

Clearly, one of the most focused ways to look for disease

genes is to use the tools of population genetics. The power

of such an approach is that it can unambiguously identify

the genes responsible for a particular disorder. However, the

more common inherited disorders are the most difficult to

study using human genetics, because combinations of dif-

ferent genes are often involved. Additional complications

can be introduced by the involvement of environmental

triggers. Genetic studies of complex disorders usually

require large study populations and can often lead to the

identification of several disease-associated loci, all of which

might require a more-detailed analysis8,9. Even if one can

identify the gene or genes associated with the disorder, such

studies provide no information about the biochemical role

of the gene products10. However, the genetic data may re-

inforce the knowledge already available, as has been the

case for Alzheimer’s disease6.

perspective REVIEWS

DDT Vol. 3, No. 4 April 1998 181

Figure 1. Positional cloning. Families displaying aclear disease phenotype are collected for geneticmapping studies. From these studies, the disease lociare positioned to chromosomal regions using geneticmarkers (a). Clones containing large inserts (such asyeast artificial chromosomes, YACS) are identified byphysical mapping (b). Candidate genes are identifiedthrough the application of transcript mapping andlarge-scale sequencing (c). The disease gene isidentified through mutation analysis, comparingaffected and unaffected populations (d).

(a)

(b)

(c)

(d)

Figure 2. Transcriptional profiling and databasemining. Oligonucleotides or cDNAs are arrayed on aglass slide (the example here shows 66 separate arraysbut many thousands are possible). The mRNA from thesamples to be compared are fluorescently labelledduring reverse transcription. The fluorescently labelledcDNA probes are then hybridized to the arrays anddifferentially expressed mRNAs are identified by thedifference of fluorescence in equivalent sequences onthe array. Grey spots represent genes expressed equally,while black and white spots represent genes that haveincreased and decreased expression levels, respectively.Alternative methods use different fluorescent labels foreach sample and a single array for the hybridizationof both samples. Differential expression is thendetected ratiometrically.

RNA from sample Areverse-transcribed to

produce a fluorescentlylabelled cDNA probe

Array probed withprobe from sample A

RNA from sample Breverse-transcribed to

produce a fluorescentlylabelled cDNA probe

Array probed withprobe from sample B

Human population studies are difficult to control, unlike

mice colonies, which are much more amenable. Recently,

several groups have used the correlation of mouse pathol-

ogy to human disease to identify putative human disease

genes11–14. Although genetic studies on mice are more effi-

cient to carry out, one still needs to demonstrate the role of

the human homologue (or orthologue) in the human dis-

order. Indeed, it is important to remember that the human

homologue of the mouse mutant might not be the mutated

gene in the human condition. Nevertheless, identification of

the mouse mutation may point the way to the biochemical

pathway involved in the disorder.

Transcriptional profiling, where the differential expres-

sion of genes is analyzed from various tissues sources (in

vitro /in vivo disease models or biopsy samples), can supply

a wealth of information regarding the biochemical processes

being modulated (see Box 1 and Figures 2 and 3). Often

these studies generate a large number of differentially

expressed genes, challenging the researcher to make selec-

tions that differentiate cause from effect. Transcriptional

profiling can be used not only to identify novel targets but

also to check the validity of a target.

Database mining for novel genes from well-characterized

structural classes, described by some as experiments ‘in silico’,

has the advantage of allowing the researcher to choose the

nature of the target being searched for (e.g. proteases or G

protein-coupled receptors). The disadvantage is that the

‘novel’ gene is often identified with no associated data on

disease relevance. Some focus can be introduced by selecting

the database to be mined; for example, a database of brain

ESTs or, better still, ESTs from specific brain regions might be

mined for homologues of genes involved in CNS disorders.

Gene identification using technologies based on function

include the yeast two-hybrid system, which has been used with

great success to clone new genes by virtue of the association of

their protein product with previously characterized proteins15.

It provides a valuable tool for identifying new genes lying

on defined biochemical pathways and for biochemically ‘map-

referencing’ genes identified by other means. Other method-

ologies for gene identification, such as expression cloning and

cloning by complementation, are also regularly used. These

approaches, like the yeast two-hybrid system, benefit from the

identification of genes within their biochemical context (e.g. as

the receptor for an orphan ligand or by substituting for another

gene product of known function)16–18.

Candidate genes will arise from all of the above

approaches, often in large numbers. In the next section, the

REVIEWS perspective


Figure 3. Samples A and B are mRNAs from tissues orcell lines (e.g. normal and diseased or drug-treated andcontrol) (a). The samples to be compared are firstdivided into 12 pools and reverse-transcribed with oneof a set of anchored 39 primers (N1 is A, T, G or C; N2 isA, G or C), containing poly(T) at their 39 end, toproduce copy DNA or cDNA (b). The products are thendivided into a number of pools (depending on the extentof analysis planned) and amplified by polmerase chainreaction (PCR) with a selection of 59 primers and theappropriate 39 primer used in the cDNA reaction. The 59primers normally comprise ten or more bases ofarbitrary sequence (represented in this example by thesequence ATGCACGTGA) (c). During the PCR, thesamples are radiolabelled with [35S]dATP or [33P]dATP toallow autoradiographic detection after polyacrylamidegel electrophoresis. Differentially expressed mRNAs areidentified by the presence or absence or change insignal on the autoradiograph (d). More recentadaptations of this technology make use of fluorescentlylabelled DNA and laser detection, which is much morequantitative than autoradiography. In addition, severalmethods have been developed that allow identificationof the individual bands on the polyacrylamide gels bytheir size and end sequence properties40.

5'-ATGCACGTGAN'1N'2TnCGCCGGCG-5'






N'1N'2TnCGCCGGCG-5'N'1N'2An-3'5'

3'

N'1N'2An-3'5'

Sample 1 Sample 2

5'-ATGCACGTGA-3'3'-N'1N'2TnCGCCGGCG-5'

Taq polymerasedNTPs

3'-N'1N'2TnCGCCGGCG-5'dNTPsMMLV reverse transcriptase

Samples A and B processed separately

(a)

(b)

(c)

(d)

sort of experimental filters that can be used to reduce these

to a number more suitable for detailed study is outlined. In

addition, many novel genes will initially be identified as

ESTs rather than full-length cDNAs, and often with limited or

no biological data. Because it is unclear what the patentable

status of ESTs is19,20, it is important to generate enough data

to secure an intellectual property position as part of the

initial validation process, and this will probably require

full-length cDNA isolation and characterization.

The gene triage processPrioritization by expression patternBefore selecting the components of a target validation

pipeline (outlined in Figure 4), it is worth stepping back and

considering what properties the ideal therapeutic target

would possess. Such a target might be a protein or nucleic

acid, the function of which can be modulated to bring about

the desired therapeutic effect with no undesirable physio-

logical consequences. To achieve this, the target protein or

nucleic acid should mediate a specific process that is directly

associated with the biochemical mechanisms causing the

disorder or its pathology. For many pharmaceutical compa-

nies, an ideal target might also function in a manner that can

be modulated by small molecules rather than requiring the

application of gene therapy, protein therapeutics or anti-

sense technology. The target validation process should con-

centrate on determining which candidate genes most closely

fit these requirements. It would seem likely that many genes

will fail to meet the above criteria as they are essential for

normal body function. Although it is certainly true that many

genes, such as the genes for the N-methyl-D-aspartate recep-

tor NR1 subunit21 and apolipoprotein B (Ref. 22), are essen-

tial for embryonic development and beyond, some genes,

such as fosB in mouse, that might have been considered

essential, or at least important, for many biochemical activ-

ities are not23. Novel and potentially interesting genes are

often identified in the absence of any definitive data con-

cerning their role, vital or not, in the adult organism. In prac-

tice, this dilemma cannot be definitively resolved for the

hundreds of genes that might be identified in a genomics

program. Under these circumstances, it is necessary to be

pragmatic. Thus, to reduce the number of genes entering the

validation pipeline, a pre-screen can be established based

on tissue expression patterns. Such a screen would not

necessarily apply to all genes. For example, genes identified

through human genetic studies should need no prioritiz-

ation by this approach; however, the data obtained will be

valuable (although lack of expression in the adult might

indicate that attention should be focused on functions

displayed earlier in embryological development). A more

perspective REVIEWS


Human genetic studiesGenes identified by this approach will play a role in the etiology of the targeted disease. However, if novel, noinformation about the biochemical role will be available(unless inferred by homology to known genes).

Transcriptional profilingThe paradigm from which transcriptionally modulatedgenes are identified will play a significant role in their valueas candidate disease genes. Well-understood in vivo dis-ease models or in vitro systems, thought to be represen-tative of the disease process, would be the most usefulsources. Additionally, the use of known disease/pathologymodifiers (pharmaceuticals or proteins) in such modelsmight well expose genes involved in therapeutic re-sponses. However, the role, if any, that such modulatedgenes may play in the etiology or pathology of the diseasewill still need to be determined.

Database mining/homology cloningUnless identified from very defined databases or libraries(e.g. regiospecific EST libraries and disease/normal sub-

tracted libraries), no information regarding the role of thesegenes in human disorders will be available at the time ofidentification. Even the use of disease or regiodefinedlibraries or databases does not guarantee identification ofgenes with a disease association. However, the advantageof this approach is that it allows the researcher to choosethe gene by the attributes of its product (e.g. presence ofstructural motifs or member of a known class of receptor).

Expression cloningGenes identified by expression cloning will already havedata associated with them; for example, they might be thereceptor for a known ligand. Further value would beobtained by the use of cDNA libraries from definedsources (tissue-specific or disease/normal subtractedlibraries).

Cloning by complementationGenes cloned by complementation already have functionaldata associated with them. As with expression cloning,increased value can be obtained through the use ofdefined libraries.

Box 2. Methods of gene identification: advantages and disadvantages

detailed analysis of expression within the tissue or organ of

interest would make use of in situ hybridization, although

the utility of in situ studies depends on the anatomy of the

target tissue. In situ data would be of great value for CNS

genes and might supply a further level of prioritization. Both

whole-body expression analysis, normally by Northern blot,

and in situ hybridization can be carried out at moderate-

throughput levels (tens to hundreds of genes) and do not

require access to full-length cDNA clones, which often

require an investment of time and resource to obtain, es-

pecially if the candidate gene was identified from database

mining. As with other areas of pharmaceutical research,

such analyses are not necessarily straightforward. The

ubiquitous expression of a gene need not exclude its role in

a tissue-specific disorder. For example, in Alzheimer’s

disease, the amyloid precursor protein is ubiquitously

expressed and cleaved to produce b-amyloid, whereas the

pathology associated with b-amyloid is

only observed in the brain24,25 (see also

Box 3). Generally, candidate genes with

expression restricted to the tissue or

organ of interest would take priority over

those having ubiquitous expression,

while the genes not expressed in the tar-

get tissue could be dropped from consid-

eration. Determining the expression

pattern through development might

also be used to assess the liabilities asso-

ciated with subsequent gene-knockout

studies and to gain some insight into

function.

The pathway a gene follows through

the validation process will vary, depend-

ing on the method by which it was

identified (Box 2). However, for genes of

unknown function, one of the main

objectives would be to identify a func-

tion at both the cellular and the whole-

organism levels. For genes identified

through homology cloning, database

mining, expression cloning, two-hybrid

cloning and complementation cloning, a

biochemical function can normally be

assigned by virtue of the cloning method.

For genes identified through human

genetic studies or transcriptional profil-

ing, even the biochemical function might

be difficult to assign. An experimental evaluation of function

requires a significant investment of resources, in terms of

time and expertise. In addition, the elucidation of biological

function will often not lead directly to the selection of the

therapeutic target. The novel gene itself may not be the best

target for therapeutic application; however, it may be the

biochemical ‘hook’ for the identification and validation of

the eventual therapeutic target. Depending on the degree of

risk aversion within the particular organization, priority

might be given to candidates that are likely therapeutic

targets in their own right – these might be described as

‘druggable’ targets (my thanks to Geoff Duyk of Exelexis

who introduced me to this useful and descriptive term).

Examples of such candidate targets would be G protein-

coupled receptors, enzymes and ion channels, and, more

often than not, these would be the focus of database mining

efforts.

REVIEWS perspective


Figure 4. New genes can enter the validation pipeline from severalsources (see Box 2). To manage the large number of genes anticipated,the pipeline should incorporate prioritization assays. The expressionpattern of the candidate genes within the whole body, and specificallywithin the organ or tissue of interest, can be used as the first level ofprioritization, because such assays have a moderate-throughput capacity.The secondary validation assays, such as functional expression andbioinformatics analysis on full-length cDNAs, will be used depending onthe particular gene under investigation, and may involve iterative cyclesbetween systems to move along biochemical pathways to the eventualscreening target. Eventually, pharmacological profiling of the candidatedrug target can be determined (if compounds are available).

New genes

Whole-body transcriptional profilingIn situ hybridization

In vitro functional expression modelsSearch for interacting gene products

Biochemical map referencing

Pharmacological profiling

Models of human pathology Analysis of clinical samples

Model animal systems

Bioinformatics analysis

Obtain full-length cDNAsGenerate antibodies

Expression and knockdown studiesOnce a candidate gene is selected for further evaluation, a

full-length cDNA clone needs to be obtained. With the

increasing need to identify and access the full coding region

of novel genes regularly and efficiently, it is becoming routine

to create high-quality cDNA libraries from the tissue sources

of interest and to grid these libraries out on an individual

clone basis, normally in 96-well plates. Gridded libraries can

be probed with ESTs or oligonucleotide sequences from can-

didate genes and full-length positive clones rapidly identified

for sequencing. In addition, the I.M.A.G.E. Consortium

(http://www-bio.llnl.gov/bbrp/image/image.html) makes all

characterized ESTs available as clones. Sometimes it is poss-

ible to use several ESTs to reconstitute a complete cDNA,

although, more often than not, the ESTs represent sequences

from the 39 end of cDNAs and might not even contain a coding

sequence. Full-length cDNAs will be required, not only for

expression analysis, antisense-oligonucleotide design and de-

signing antipeptide antibodies for in vitro and in vivo studies,

but also for further structural analysis based on the full protein

sequence. Because the analysis of protein expression, both in

vitro and in vivo, is an essential part of functional evaluation,

it is important to begin generating immunological reagents as

soon as possible. Indeed, when using antisense knockdown

as a tool for elucidating gene function, it is critical that the

reagents are available to demonstrate reduction in protein

expression. These reagents can be made using peptides, or

through recombinant proteins expressed in bacteria. In addi-

tion, expression constructs can be made using epitope

tags26–28. This is a valuable way of rapidly tracking the expres-

sion of proteins in cells, without the need to generate specific

antibodies. It should be noted, however, that the tags might

affect protein function and/or subcellular location. This con-

cern would be especially acute with novel proteins having no

known function. Immunological reagents produced against

the unmodified gene product will enable tissue and subcellu-

lar protein expression to be determined immunohistochemi-

cally with more confidence (this is especially important in the

brain, where the sites of protein synthesis and functional

expression may be distant). In addition, these reagents will be

important for in vitro expression studies. In parallel with anti-

body production, efforts to express the candidate gene in cell

culture should be initiated, because a significant amount of

data can be gained from in vitro expression studies in mam-

malian cells. For proteins having a potential role in signal

transduction, expression systems can be designed to identify

the pathways these proteins might affect. These systems are

available commercially and comprise the various components

required to reconstruct known signal transduction pathways

and report their activity in cell culture29. Transient transfection

assays may not be sensitive enough for phenotype analysis if

the biochemical function of the protein is subtle. Creating per-

manent cell lines for expression of the candidate protein gen-

erates a tool with significantly more utility, but takes time to

achieve. Often it is wise to embark on transient studies,

because they require a limited time commitment.

When planning in vitro expression studies, the following

questions should be considered:

• Based on the candidate gene under evaluation, should a

specific cell host be selected (such as neurone,

haemopoietic cell or vascular endothelial cell)?

• Is expression of the candidate gene likely to be detri-

mental to the cell during long-term culture?

• Is it possible that the host cell line displays endogenous

expression of the candidate gene?

Many potential host cell lines are available from the ATCC

(American Type Culture Collection; http://www.atcc.org/

atcc.html) and these might make useful starting points. If

the candidate gene shows activity in the in vitro expression

perspective REVIEWS


Box 3. A gene for epilepsy

The identification of disease genes using linkage studiesrequires no knowledge of the disease gene function.However, given the significant amount of workrequired to identify such a gene, it is always hoped thatits sequence might provide some clue as to its role inthe disease. Unfortunately, this is not always the case.Recently, researchers in the USA and Finland havecloned the gene for progressive myoclonus epilepsy(EPM1)49; this disorder is a heterogeneous group ofsevere inherited epilepsies, characterized by myoclonicseizures, generalized epilepsy and progressive neuro-logical deterioration. They found the gene coding forcystatin B, which is a cysteine protease inhibitor50,51,and proposed that cystatin B might play a role in inacti-vating proteases leaking from lysosomes. Cystatin B isubiquitously expressed, although EPM1 appears to bethe only phenotype it produces in affected populations.This unexpected finding (i.e. genes directly involvedwith synaptic transmission were not identified) has ledto speculation that there may be a functional common-ality between lysosomes and synaptic vesicles52.However, more work is required to determine the rolecystatin B plays in progressive myoclonus epilepsy.

system (which would be the hope), this might lead to tox-

icity. If a permanent cell line is the eventual aim, it would be

worth considering the use of an inducible expression system

to allow control over when the gene is expressed. Recently,

two inducible systems have been developed that not only

give low background expression (mammalian inducible sys-

tems have been notoriously leaky for expression from so-

called inducible vectors) but can also be used in a

dose–response mode with the inducing agents, allowing

some control of expression levels30,31. In addition, both sys-

tems can be used in transgenic animals. If the gene of inter-

est is found to be endogenously expressed in cultured cells,

consideration should be given to examining the effects of

knockdown studies using antisense oligonucleotides. In cell

culture, antisense sequences can be delivered as oligonu-

cleotides32 or even in antisense expression constructs. The

combination of antisense and overexpression in vitro may

be highly informative because the analysis of activity over a

range of expression levels is possible.

If the function of the gene of interest is known, and es-

pecially if it represents a known class of pharmaceutical tar-

gets, expression in mammalian cells will enable a rapid

determination of its pharmacological properties. Expression

of the candidate genes in cell culture will often be the first

opportunity for a pharmaceutical company to make use of

its compound library (arguably its most valuable resource)

in the target validation process. The most obvious value for

screening a selection of proprietary, as well as public

domain compounds, would be to evaluate the pharmaco-

logical profile of candidate genes from database mining

efforts, as these would have been selected based on their

homology to known pharmacological targets. In addition,

one could also evaluate the effect novel genes had on

known targets or biochemical pathways through the modifi-

cation of the pharmacology (e.g. altering desensitization of

G protein-coupled receptors). Systems such as PathDetect29

and other reporter systems would be valuable in this regard.

Finding pharmacological tools early in the evaluation

process will enable rapid in vivo analysis of function and

may also aid in the construction of disease models.

Animal modelsThe functional evaluation of many novel genes will not

easily succumb to in vitro analysis, either because they only

function in specialized cell systems or because they require

complex multicellular systems to reveal the phenotype. In

these cases, functional analysis will require the application

of animal model systems (Box 4). The choice of animal

model will depend on the expected biology of the candidate

gene under consideration.

The use of less complex organisms such as yeast, worms

or flies will, in many cases, be valuable in determining func-

tion, especially for genes whose function is required during

development. However, the determination of function will

not necessarily lead to the identification of the appropriate

therapeutic target. For this, a mammalian animal model will

be required. Several mammalian species can be used for

transgenic models of gene overexpression, but at present

only the mouse is available as a model for gene knockouts

or mutagenesis via homologous recombination33. Many natu-

ral mouse mutant strains have been described34 and these

might provide a valuable resource for disease models asso-

ciated with novel genes, in addition to being the route for

new gene discovery. However, the dilemma remains that a

mouse model of a disorder cannot predict the human dis-

ease process with certainty. In addition, the use of animal

models that were developed and ‘validated’ using clinically

efficacious compounds might not be appropriate for the

novel targets that emerge from genome research. Whenever

possible, reference to human disease biology should be

made either through the availability of surrogate markers or

through the use of human polymorphism detection – not

only of the genes coding for the potential therapeutic target,

but also of genes whose products fall on the same bio-

chemical pathways. As more data on mapped genes become

available, the latter approach should become more feasible.

Some targets, such as those involved in disorders of behav-

iour, will be harder if not impossible to validate in animals

through the use of surrogate markers, as they probably

involve subtle alterations in neurochemical pathways35 and

may necessitate a human genetics approach.

ConclusionThis review has attempted to give an indication of the chal-

lenge the pharmaceutical industry is facing in turning the

information content of the human genome into therapeuti-

cally valuable and viable commodities. The magnitude of

this challenge is clearly illustrated when we consider that, to

date, this industry has produced drugs acting against just

over 400 human molecular targets36, and, with screening

and combinatorial chemistry technologies increasing the

efficiency of the discovery process, the demand for suitable

targets is rapidly outstripping the supply. It has been

estimated that between 3,000 and 10,000 new targets for

REVIEWS perspective


therapeutic intervention will emerge from genomic

research36, and it is these new targets that will fill discovery

pipelines in the future. It is likely that the limiting phase of

the drug discovery process will soon become the validation

of the therapeutic target. This process therefore needs to be

made as efficient as possible, not only to fuel the discovery/

development engine, but also to ensure a solid proprietary

position for each new target. Clearly, many technologies,

some of which are not normally associated with pharma-

ceutical research, will be required to characterize and

develop the new therapeutic targets, and it is likely that no

single organization will be able, or even want, to develop all

perspective REVIEWS


It is fortunate that it is not only the human genome that issubjected to intense sequencing efforts. The genomes ofseveral model organisms are presently being sequenced.For Saccharomyces cerevisiae it has been completed53

and the sequence of the nematode worm Caenorhabditiselegans is >50% complete. At present there is limited coverage of other model organisms such as the fruit fly,Drosophila melanogaster, and the mouse.The value of model organisms such as the mouse requireslittle justification. As a mammal, the mouse is closelyrelated to humans and displays many pathologies similarto those of humans, obesity being a good example11,12,14.The value of lower animals relates more to their utility inthe determination of gene function rather than pathology.How yeast and C. elegans aid the functional characteriz-ation of novel genes is described below. However, othermodel organisms, such as D. melanogaster and the pufferfish, Fugu rubripes, are also likely to play significant rolesin the functional characterization of novel genes54,55.

Single-cell systemsYeasts are single-celled eukaryotic organisms that arealmost as easy to culture and genetically manipulate asEscherichia coli. The genome comprises 13,105,020 bparranged on 16 chromosomes56. Considering the evolu-tionary separation between humans and yeast, it is strikingthat so many yeast genes are homologous to humangenes57. Determining gene function in yeast is compara-tively easy, so it is anticipated, based on the knowledge ofits genome sequence and its facile genetics, that yeastwill be a productive tool for a subset of therapeutic targets,including genes involved in tumorigenesis or hyperplasiawhere mutagenesis followed by clonal amplification ofteninitiates the pathology.

Multicellular systemsMany genes display their function only within the contextof a multicellular system, or at a specific point in the devel-opmental process. Therefore, well-characterized multicel-lular model organisms are likely to become valuable toolsfor the dissection of gene function. C. elegans has beenused as a model organism to analyze the genetics ofdevelopment since its utility was first recognized bySidney Brenner in the 1960s. The adult organism contains959 cells, and the origin of each of these cells has beentracked through the developmental process58. Rather like

yeast, many C. elegans genes show functional homologyto mammalian genes. An example of where our under-standing of a mammalian system has been greatlyenhanced by studying C. elegans is the phenomenon ofprogrammed cell death, or apoptosis59.Apoptosis is used to remove unwanted or damaged cellsin an orderly fashion. It is a vital process for the normaldevelopment of multicellular organisms and is likely to playa role in disorders such as Alzheimer’s disease andParkinson’s disease60. Mutations in 14 different geneshave been shown to affect apoptosis in C. elegans. Mostare involved in the death of small cell populations; how-ever, three genes (ced-3, ced-4 and ced-9) are involved inthe ’death’ decisions of all cells; ced-3 and ced-4 are pro-apoptotic, while ced-9 is anti-apoptotic. Genetic studiesusing C. elegans double mutants have demonstrated theepistatic relationship between these genes – ced-9 func-tioning upstream of ced-3 and ced-4; ced-9 is homologousto the mammalian anti-apoptotic Bcl2 gene and ced-3 ishomologous to the cysteine protease family of genes61.However, until recently no ced-4 homologue had beenidentified in mammalian cells. It has been shown that theced-4 gene product can interact with both Bcl-xL (a mem-ber of the Bcl-2 family) and interleukin-1b convertingenzyme (ICE, a member of the cysteine protease family)62.It is clear, based on these observations, that a mammalianced-4 functional homologue exists that biochemically couples the activities of the cysteine protease and Bcl-2family of proteins.Transgenic mouse models have been available to theresearch community for some time33,63. The value of theseanimals lies in their use both for the analysis of gene function and for the creation of disease models. However,the use of mouse transgenic models can be limited by the time it takes to generate the appropriate strains andother factors, such as regiospecific expression of thetransgene, as well as the impact of expression or geneknockout on the developmental process. The use of anti-sense technology in the context of gene ‘knockdown’ hasthe potential to overcome some of these problems64. Inaddition, antisense oligonucleotides designed and vali-dated for in vitro studies can be used in vivo to examinegene knockdown effects rapidly. Although effectiveoligonucleotide design has been a problem, new empiricalmethods of antisense oligonucleotide selection arebecoming available32.

Box 4. The use of animal models for gene function studies

these in-house. This is clearly illustrated by considering the

unpredictability of the validation process that is required for

genes identified through human genetic studies. The role

these genes play in the disease process can be obscure, and

it will be impossible to predict what the validation pathway

will be. For this reason, alliances between major pharma-

ceutical organizations and genome-based biotechnology

companies based on specific technologies, rather than

therapeutic focus, are likely to become more common. It is

also likely that multiple smaller collaborations or alliances

will be required to facilitate the efficient discovery and

development of the next generation of pharmaceuticals. A

model for collaborations that might gain favour in the future,

as it would allow the maximal use of cutting edge technol-

ogies, would be the linear organization of alliances for gene

discovery, target validation and ultra-high-throughput

screening coupled with combinatorial chemistry.

The industry must also face the fact that novelty, by

its very nature, creates a degree of uncertainty as compounds

against these new targets move towards the clinic. Some of

the alternative approaches for gene discovery discussed

above offer the advantage of target-type selection, either up-

front, as in database mining, or through prioritization, as in

transcriptional profiling. However, the role these genes

might play in the human disease process may eventually

require validation in the clinic. The industry will have to rely

more heavily on novel animal models of disease that cannot

be ‘back-validated’ with known drugs. However, genomics

research offers the opportunity of streamlining the develop-

ment process, by increasing the tools available for drug

metabolism and toxicology studies, as well as patient selec-

tion for clinical trials. Genomics is therefore set to change all

aspects of the drug discovery and development process.

AcknowledgementsI thank my colleagues at Wyeth-Ayerst Research, especially

Jim Barrett, Maria Betty, Mark Cockett, Andrew Wood and Ken

Rhodes, for their contributions to the preparation of this article.

REFERENCES1 Antequera, F. and Bird, A. (1994) Nat. Genet. 8, 114

2 Fields, C. et al. (1994) Nat. Genet. 7, 345–346

3 Schuler, G.D. et al. (1996) Science 274, 540–546

4 Papadopoulos, N. et al. (1994) Science 263, 1625–1629

5 Hillier, L.D. et al. (1996) Genet. Res. 6, 807–828

6 Selkoe, D.J. (1997) Science 275, 630–631

7 Collins, F.S. (1995) Nat. Genet. 9, 347–350

8 Murphy, K.C., Cardno, A.G. and McGuffin, P. (1996) J. Mol. Neurol. 7, 147–157

9 Pulver, A.E. et al. (1995) Am. J. Med. Genet. 60, 252–260

10 Roelfsema, J.H. and Breuning, M.H. (1996) Adv. Nephrol. Necker Hosp. 25, 131–145

11 Fan, W. et al. (1997) Nature 385, 165–168

12 Lee, G.H. et al. (1996) Nature 379, 632–635

13 Tartaglia, L.A. et al. (1995) Cell 83, 1263–1271

14 Zhang, Y. et al. (1994) Nature 372, 425–432

15 Fields, S. and Song, O. (1989) Nature 340, 245–246

16 Ko, C.H. and Gaber, R.F. (1991) Mol. Cell. Biol. 11, 4266–4273

17 Seed, B. (1995) Curr. Opin. Biotechnol. 6, 567–573

18 Simonsen, H. and Lodish, H.F. (1994) Trends Pharmacol. Sci. 15, 437–441

19 Auth, D.R. (1997) Nat. Biotechnol. 15, 911–912

20 Eisenberg, R.S. (1996) Trends Biotechnol. 14, 302–307

21 Forrest, D. et al. (1994) Neuron 13, 325–338

22 Farese, R.V., Jr et al. (1995) Proc. Natl. Acad. Sci. U. S. A. 92, 1774–1778

23 Brown, J.R. et al. (1996) Cell 86, 297–309

24 Selkoe, D.J. (1991) Neuron 6, 487–498

25 Selkoe, D.J. (1993) Trends Neurosci. 16, 403–409

26 Kolodziej, P.A. and Young, R.A. (1991) Methods Enzymol. 194, 508–519

27 Wang, L.F. et al. (1996) Gene 169, 53–58

28 Sells, M.A. and Chernoff, J. (1995) Gene 152, 187–189

29 Xu, L., Sanchez, T. and Zheng, C-F. (1997) Strategies 10, 1–3

30 No, D., Yao, T.P. and Evans, R.M. (1996) Proc. Natl. Acad. Sci. U. S. A. 93, 3346–3351

31 Gossen, M. et al. (1995) Science 268, 1766–1769

32 Ho, S.P. et al. (1996) Nucleic Acids Res. 24, 1901–1907

33 Jacobson, D. and Anagnostopoulos, A. (1996) Trends Genet. 12, 117–118

34 Darling, S. (1996) Curr. Opin. Genet. Dev. 6, 289–294

35 Campbell, I.L. and Gold, L.H. (1996) Mol. Psychiatry 1, 105–120

36 Drews, J. (1996) Nat. Biotechnol. 14, 1516–1517

37 Schena, M. et al. (1995) Science 270, 467–470

38 Schena, M. et al. (1996) Proc. Natl. Acad. Sci. U. S. A. 93, 10614–10619

39 Liang, P. and Pardee, A.B. (1992) Science 257, 967–971

40 Prashar, Y. and Weissman, S.M. (1996) Proc. Natl. Acad. Sci. U. S. A. 93, 659–663

41 Lockhart, D.J. et al. (1996) Nat. Biotechnol. 14, 1675–1680

42 DeRisi, J. et al. (1996) Nat. Genet. 14, 457–460

43 Fickett, J.W. (1996) Trends Genet. 12, 316–320

44 Bork, P. and Bairoch, A. (1996) Trends Genet. 12, 425–427

45 Boguski, M.S. (1995) New Engl. J. Med. 333, 645–647

46 Savitsky, K. et al. (1995) Hum. Mol. Genet. 4, 2025–2032

47 Savitsky, K. et al. (1995) Science 268, 1749–1753

48 Watson, S. and Arkinstall, S. (1994) The G-Protein Linked Receptor Facts Book,

Academic Press

49 Pennacchio, L.A. et al. (1996) Science 271, 1731–1734

50 Turk, V. and Bode, W. (1991) FEBS Lett. 285, 213–219

51 Jarvinen, M. and Rinne, A. (1982) Biochim. Biophys. Acta 708, 210–217

52 O’Brien, C. (1997) Science 271, 1997

53 Mewes, H.W. et al. (1997) Nature 387 (Suppl. 6632s), 7–65

54 Osborne, K.A. et al. (1997) Science 277, 834–836

55 Brenner, S. et al. (1993) Nature 366, 265–268

56 Goffeau, A. et al. (1996) Science 274, 546 and 563–567

57 Bassett, D.E., Boguski, M.S. and Hieter, P. (1996) Nature 379, 589–590

58 Sulston, J.E. et al. (1983) Dev. Biol. 100, 64–119

59 Hengartner, M.O. (1996) Curr. Opin. Genet. Dev. 6, 34–38

60 Spence, P. et al. (1996) Expert Opin. Ther. Pat. 6, 345–366

61 Steller, H. (1995) Science 267, 1445–1449

62 Chinnaiyan, A.M. et al. (1997) Science 275, 1122–1126

63 Gordon, J.W. et al. (1980) Proc. Natl. Acad. Sci. U. S. A. 77, 7380–7384

64 Christoffersen, R.E. (1997) Nat. Biotechnol. 15, 483–484

REVIEWS perspective


Documents

Obtaining value from the human genome: a challenge for the pharmaceutical industry